Thursday, March 2, 2023

Mark Goldstein : It’s Alive! · On A.I. Generated Poetry

Imagination is the only weapon in the war against technology.[1]
after Jules de Gaultier



From the porta of capitalism, we are presented with yet another dejection, this time in the actuality of Artificial Intelligence via ChatGPT. The hubris inherent in this vendable tech is ungodly, and is a pure symptom of the “interesting” times in which we now live. As ever, anyone who is a professional writer and reader aims to keep abreast of the very best that our culture has produced over millennia. From the clay tablets at Sumer, up through Egyptian papyri, through to woodblock printing in ancient China, on to Gutenberg and his press, and to the advent of the personal computer — all have produced works of literary merit. New to this fray is A.I. tech as a generator of written work. Fabricated by those with a near total illiteracy, and likely driven by some desire to vanquish the artist-seer/story-teller from the planet, we are made to believe that this technology is benign and, hence, ready for use, royalty-free. What might this mean for the writer, and specifically for the poet, playwright, or novelist? Might A.I. technology spell the end of the independent creative artist as midwife to literary culture?

In the December 16, 2022 edition of the New York Times, in an opinion piece entitled, “ChatGPT Has a Devastating Sense of Humor,” Farhad Manjoo writes: “I started talking to ChatGPT a couple of weeks ago, after the artificial intelligence company OpenAI released the bot as a ‘research preview’ of its work on large language models. A language model is an A.I. system that has been trained on enormous troves of text to find the probabilistic connection between words; ChatGPT is a language model that has been optimized to create what’s long been the holy grail in artificial intelligence research — a computer with which you can hold a conversation.”[2]

Manjoo’s opinion piece is devoid of any hard ethical concerns regarding A.I. Instead, consistent with the current push for tech’s cultural credibility and widespread acceptance, Manjoo takes a neutral approach assuring us that “OpenAI’s chief technology officer told [him] the company is carefully monitoring how people use and misuse it, quickly altering the system to address evident harms and iteratively improving it in response to user feedback.” This laissez-faire attitude is prevalent throughout both news and social media, with the exception of a handful of artists (mostly visual) who are questioning whether A.I. may signal the end of their place in the zeitgeist. This quest for the “holy grail” in A.I., or in anything for that matter has, in the past, led us into real trouble as a society. One need call to mind the Manhattan Project and its leader, physicist Robert Oppenheimer who, famously, said of his having helped create the atomic bomb (quoting the Bhagavad Gita): “Now I am become Death, the destroyer of worlds.”

The severe limitation of A.I. technology may best be expressed in what is known as the Chinese Room argument. “The argument holds that a digital computer executing a program cannot have a ‘mind,’ ‘understanding’ or ‘consciousness,’ regardless of how intelligently or human-like the program may make the computer behave. The argument was presented by philosopher John Searle in his paper, ‘Minds, Brains, and Programs,’ published in Behavioral and Brain Sciences in 1980.”[3]

“The Chinese Room argument is directed against the philosophical positions of functionalism and computationalism, which hold that the mind may be viewed as an information-processing system operating on formal symbols, and that simulation of a given mental state is sufficient for its presence. Specifically, the argument is intended to refute a position Searle calls strong A.I.: ‘The appropriately programmed computer with the right inputs and outputs would thereby have a mind in exactly the same sense human beings have minds.’”[4]

In light of the recent ChatGPT public relations campaign, the above argument may not offer much solace. On the other hand, if sentience is unlikely, might A.I. simply fail and just go away? This too is unlikely, given that marketing strategies of similar technologies have been tried and failed before, yet the technologies are still with us. In 1984 The Policeman’s Beard is Half Constructed: Computer prose and poetry by Racter was published by Warner Books.[5] The Policeman’s Beard is “the first algorithmically-authored book to be printed and marketed towards a mass readership. It has garnered a cult following amongst experimental literature enthusiasts. It is notoriously difficult to acquire a physical copy of the book, and the mystique surrounding Racter’s functionality remains as strong as ever.” “Racter’s legacy isn’t one of a technological break-through; the publication of The Policeman’s Beard and the subsequent release of the Racter computer program represent important moments in the history of algorithmic authorship wherein computer-generated texts became marketed for mass consumption.”[6]

“Marketed for mass consumption”: precisely. In light of this current effort, where are we now, and where is the writer in all of this? Let’s return to the Times piece, wherein we’re told that “A language model is an A.I. system that has been trained on enormous troves of text to find the probabilistic connection between words.” Trained, how? As if it were a pet? Which texts? Those that constitute a literature? If so, whose? ChatGPT is a new version of A.I. technology, one in which Microsoft is currently investing 10 billion dollars to further develop. So there’s a lot at stake here. Hence, it’s important that we keep in mind that Manjoo’s dispatch from the Times is propaganda, meant to sell us on the benefits of a soon-to-be monetized technology and to push many to pursue ambitions that they might not otherwise entertain, such as those of “poet,” “novelist” or “playwright.”


What is it to imagine? How does the imagination itself underpin our work as writers, and as poets? For William Blake, the imagination is not a state: “it is the human existence itself.” Therefore, I am sure that the imagination resides, to a varying degree, within each and every living human being. In some aspect of the term, imagination may even be shared between us.

Consider the typo, the accidental, the error along with the misunderstood or even the irrational as elements of the poetic imagination. Specifically, with regard to ChatGPT, eureka does not equal “troves” of data. Poet Wendell Berry said, “Every day do something that does not compute.” This “irrational” practice is but one approach that remains central to the poet’s work. How else is the writer to escape the obvious and formulaic aspects of the language? Hannah Arendt said that English as it is now written is little more than an endless string of clichés, and so the poet’s effort must be redoubled. How else to imagine?

It is worth noting that some of our greatest poetry involves juxtaposing the so-called wrong words in the right way — at this time, is A.I. capable of making this crucial, aesthetic leap? Has ChatGPT been encoded by those who understand the art of poetry or of literature? Or are they unknowingly working against this spirit of experimentation, of imagination? In the poetry samples generated by ChatGPT are we to find a blind adherence to a conservative approach? Will ChatGPT’s texts suffer from a proliferation of cliché underpinned by a narrow methodology to what may constitute a poetry? In short, is its written work the work of the amateur? If so, then perhaps we should accept ChatGPT into our community? Then, it too can conduct its A.I. education in public, as we the reader-writer-critic encounter its debased poetry, botched novels, and trite short stories.


In light of this technology’s public “education,” let’s interrogate ChatGPT directly, for it appears that, self-admittedly, A.I. cannot imagine. Odd, seeing as another A.I. technology, Midjourney,[7] uses “imagine” as its key search term. However, this is beside the point. With regard to ChatGPT, let’s put the technology through its paces with a specific focus on determining its limitations via the use of contemporary poetry.

Noam Chomsky, in his former role as professor of linguistics, reminds us that poetry is the highest form of creative expression in language. In response to this specific observation I fed ChatGPT prompts that were poetically based. I asked it to write a poem in the style of the late, and hugely influential, American poet and artist, Joe Brainard.[8] Here’s what it generated:

“Try again,” indeed. The above is a perversion of a poem by Brainard. Sure, it has a few minor hallmarks of his style, and even some of his language, such as the anaphora, “I remember” but this is little more than an aping through paraphrase, using old models of what may constitute “poetry.” It is not Brainard-esque poetry either for it ignores Brainard’s form, which is prose poem built in modules of long lines constituting individual units. These hallmarks are emblematic of the strong influence of Brainard’s precursors in the New York School as well as the Beats — specifically, Frank O’Hara and Allen Ginsberg. His masterwork, I Remember, has been described thus: “Painterly in its vivid details and collagist in its hands-off juxtaposition, it is an accumulative, oblique biography, a portrait of the artist as a young man. It is much, much greater than the mere sum of its parts. [John] Ashbery referred to it, only half-jokingly, as ‘humane smut.’ It has that sweet, playful self-possession that pervades Brainard’s work.”[9] Of Brainard’s book, Paul Auster has written that, “I Remember is a masterpiece. One by one, the so-called important books of our time will be forgotten, but Joe Brainard's modest little gem will endure. In simple, forthright, declarative sentences, he charts the map of the human soul and permanently alters the way we look at the world. I Remember is both uproariously funny and deeply moving. It is also one of the few totally original books I have ever read.”[10] This is key to the work’s form, which is, as Auster states correctly, forthright, built of declarative sentences, and seemingly simple, yet ChatGPT cannot approximate its literal, or figurative, value. Why is this? In light of such shortcomings of ChatGPT, it’s important to note that Ezra Pound’s maxim, “Make it New” (now, almost a cliché among poets), remains a very useful key when reading any or all poetry. Specifically, this newness that we seek in poetry is a braiding of form and content. The poem itself is an embodiment of this braid. It is in light of such caveats around poetic creation that ChatGPT’s limitations become better defined.

Next, I asked ChatGPT to write a poem in the style of the somewhat obscure, Pulitzer Prize-winning poet, George Oppen.[11] Here’s what it produced:

Aside from the obvious hiccups, again, the poem is awful, and bears little resemblance to the unique style of one written by Objectivist poet Oppen. The poem’s muddled rhyme scheme is an abomination of Oppen’s earliest published poetry. Might ChatGPT’s result be a shortcoming of my ability to write the necessary “prompts” required to generate said poetry? Or is ChatGPT too inchoate to produce such a work? More generally, why is ChatGPT’s A.I. technology unable to aggregate poetic material into a poem that is “original,” “meaningful” and “new”? Perhaps, in ChatGPT’s “borrowing” from an available online pool of works by Oppen, it is running into many examples that have little bearing on the highest artistry that, in his lifetime, he was able to achieve? Finally, perhaps those that programmed ChatGPT are altogether unaware of Oppen’s poetry?

In part, written poetry is created out of feeling. Sure, this feeling may be one of indifference, ambiguity, bitterness, hatred, love, rage, etc. But feeling is, in part, at the root of it, and feeling precedes language. When we are in utero, we feel. We don’t, as yet, have language per se and, hence, we are in a sense, pure awareness. The algorithm does not feel. It cannot be biologically pre-linguistic. It does not have these layers of development, these strata, in a sense, from which to mine the material of poetry. The algorithm, the machine, cannot feel. And so, it produces an attempt at poetry that apes feeling, at best, and at worst is a kind of mimicry or cliché.

Poetry is difficult to write well. So much contemporary poetry is garbage. Perhaps ten per cent of it is worth reading at all, and of that ten per cent, perhaps one per cent will outlast the culture in which it is written. It is with this in mind that I asked ChatGPT to write a poem, this time in the style of award-winning poet Phil Hall.[12] Here’s what it produced:

The poem that ChatGPT “writes” is nothing at all like a poem by Hall. Formally, again we’re presented with tercets or quatrains with rhymes that are slightly less obvious than those that appear in the “Oppen” poem. Yet, aside from it drawing on common tropes found in the pastoral, and that its subject is the poet in the natural world, there’s little to tie this generated text to the body of work published under the name, Phil Hall. This raises the question, what exactly is a poem by Hall like?

rob mclennan says of Hall that “Unlike poets such as Al Purdy or George Bowering, Phil Hall’s work makes it impossible to excise any obvious series or sequence of ‘greatest hits’—i.e.: poems that readers already know and love re-packaged for the general reader out of his numerous books and chapbooks; instead, what becomes compelling about Hall’s work is […] how many permutations of [it] might [be] possible, as well as the remarkable durability of his writing. He writes on influences, literary and otherwise; he writes of rural Ontario, and such terrible violence even — as he processes such; he stitches together lines and tall tales and fables from his life and the stories that float around the ethos of his variety of Ontario wilds.”[13]

mclennan’s comments raise a critical question with regard to A.I. technology and ChatGPT, specifically: is Hall’s style, or any style that is based on a unique arrangement of language drawn from disparate materials ape-able without access to such materials? Can a unique and complex style be mimicked meaningfully? Of course, our literature is full of hoaxes and parodies that have unexpectedly produced meaningful works: James McAuley’s & Harold Stewart’s The Ern Malley edition of Angry Penguins; Kenneth Rexroth’s The Love Poems of Marichiko; and Kent Johnson’s Doubled Flowering · from the Notebooks of Araki Yasusada. Despite these authors’ intention to deceive, all of these works were written by professional poets drawing on an incalculable set of materials as underpinning to these works.

Is a poem a unique occurrence in language, one that fluctuates from poem to poem? If so, then what is there to mimic? What of the serial- or long-poem? Joe Brainard said that style in a particular work is everything that remains despite its author having hoped to avoid it. Perhaps this is ground for all writing? Hence, when drawing on an original work can ChatGPT hope, and if so, can it hope for parody — for in order for a work to be original shouldn’t it avoid sounding too similar to other poetry?[14]

As I began to sense some of the deeper limitations of ChatGPT, I asked it to produce poems in imitation of two of the most highly regarded poets the Western world has produced: Gertrude Stein (even going so far as to specify Stein’s canonical masterwork from 1914, Tender Buttons), and John Ashbery.

Again, ChatGPT produces terrible poems that look like poems and sound like poems (mostly from the late 19th century) but bear little or no relation to the work of the authors suggested by the prompts. I suspect that ChapGPT has a very limited sense of form and of the formal urgencies/possibilities at work in much of our lasting poetries.

I can’t help but wonder if, as an artist, perhaps one’s greatest fear may be cheap imitation, which can accelerate one’s own sense of impermanence? Of course, this implies some notion of permanence, of which there is none. For example if we look at the body of work by Aeschylus, the father of ancient Greek drama, we may note that he wrote between seventy and ninety plays, of which only seven survive.

As a creative translator of poetry into English, over decades I have used A.I. technology to aid in my process of translation. This methodology has at times employed a number of different online translators so as to generate variant translations of specific words, lines, or texts. Ultimately, these variants must be sorted and combined into a final text based on some intuitive sense of aesthetic value. It is in this work that taste comes into play. American composer, writer, artist and poet, John Cage — a devout adherent of the aleatoric, or chance-based, approach to text generation — admitted that in his mesostic poetry, wing-words were often “determined by taste.”

It is this — the artists ability to determine by taste — that is key to any potentially powerful work. Unsurprisingly or not, it seems that A.I. technology is lacking in taste, in acculturation, and in the ability to synthesize and interpret beyond the given, beyond the written, and beyond the seen.

So, what does this tell us about ChatGPT and A.I. tech in general? Likely, that poetry — at least, masterful poetry of the highest order — is, for now, beyond the reach of A.I. word generators. Moreover, the spectre of the Chinese Room argument as it pertains to weak A.I. tech and its writing of poetry seems to take on an added dimension — without what the Gnostics called the God spark, there can be no real poetry. This development is momentarily encouraging, seeing as the aping of styles, whether through A.I. or in the real, has allowed for much bad work to be written. Yet one cannot help but recognize that this particular piece of tech is in its earliest stages of development. This being the case, I decided to ask ChatGPT to speak to itself directly:

There it is: mimicry based on patterns and rules that ChatGPT has been trained on. Of course, a number of critical questions remain: who or what is driving this technology, and why? Let’s return to subscription-based A.I. software, in general, so as to pinpoint where exactly we now stand.


The development of this technology is driven and monetized by individuals who have seldom, if ever, allowed themselves access to the raw creative forces that have produced someone like a James Joyce. There’s no doubt that Joyce himself would be the first to admit that his masterworks took years of dedication to create and, even then, that they were the result of a lifetime’s commitment to his discipline. Without this level of commitment, there is a real chance that Joyce might not have been able to see beyond his current culture’s artistic limitations. He may have done little more than mimic what was in the common air.

Certainly, there’s a place in our culture for A.I. rendered, bot-driven poetry — for those who don’t have the time, education, or means to do the long careful work of reading through much of the material required to understand what a poem is and how it operates. Isn’t this the target audience for A.I. writing? Isn’t A.I. the great leveller against the cultural gatekeeping posed by the publishing industry, academia, and the vast communities of active poets, writers and artists — all of whom seem to have their hands on the levers of culture? So, what if these bot poems will further congest these mechanisms, those which are already choked with millions of unread pieces of writing created by amateur and professional poets alike? Yes, poetry has been sidelined in much of contemporary culture, where it is an almost non-existent force. In this air, there is room only for a few token poets. It’s here that we find “bestselling author-poets” such as actor James Franco, or singer-songwriter Jewel. Now, through A.I. technology, these future Rod McKuens may be joined by other soon-to-be best-selling poets. For poetry is about sales, is it not? And A.I. technology can facilitate such commerce for a subscription at a mere number of dollars per month. The dream of being a widely-read cultural force, a kind of demi-god immortal in literature is now within reach for all. Or is it?

It seems obvious that there’s an element of envy driving the development of A.I. technology. We live in a culture of the self, one that is enormously egocentric and requires instantaneous gratification. The idea that a lifetime is limited, and that this limitation also bounds one’s ability to do only so much with the time that we are given is frustrating to anyone who has seriously studied the craft of writing, or any art form whatsoever. Yet this reality is anathema to A.I. tech, which perceives itself to be deathless. Immortality is one of the goals here, as pushback against these life-limits on our craft. For as living beings who are also writers, we must ever contend with Chaucer’s translation of Hippocrates’s famous aphorism “Ars longa, vita brevis”: “The lyf so short, the craft so long to lerne.”

A.I. developers push against this insight, and monetize the machine’s ability to generate “well-written” work in any genre while eschewing the necessary time to do so … forever. This great “liberation” from technique, and from the body, might allow everyone to write, for example, a novel, despite lacking the ability to do so. One wonders if A.I. might come to write a series of works over many human lifetimes? One that could be sold to generations of readers? A kind of serialized story that parallels its creator’s own immortality? The thinking goes that A.I. may finally bridge this creative gap.

What gap? Without imagination, discipline, talent, and luck, it is likely that we will only ever be re-presented with current cultural models in infinite variation but with no real relevance to thought, or experience, or contemporary living.

Poetry written by coders and processed by a computer reads like poetry written by coders and processed by a computer. Worse, at the outset the coder’s knowledge of poetry parallels the poet’s knowledge of coding, which is to say that both are lost in their secondary fields. Moreover, as in the Chinese Room argument regarding A.I., there is no seeing beyond when the beyond itself is yet to be imagined. A.I. cannot do this for us. We, as human beings, must do this for ourselves: we must imagine. For this we need visionaries, artists unafraid to write into the darkness at the edge of the known. For A.I., there is no darkness, only data “scraping” from the digital realm. A realm that is becoming more limited in what it can perceive. For A.I. there is no body, there is no birth and death. Freedom from the known, this is what our culture — or any culture that is unassailably alive — requires. For this we must turn to the flesh-and-blood reports of the living-dying artist. For as one of our greatest artists, Franz Kafka, has written, “the meaning of life is that it stops.”

With regard to the poem, at this point in A.I.’s development it is unlikely the technology will put the poet out of work. Since A.I. does not know death, it certainly does not know life, and then what is left for it to write about? There is little doubt that the contemporary poet, or for that matter, poetry itself, is marginal to the culture; yet in this marginality it is essential too. For it is from the margins, from this unique cultural viewpoint, that it is able to see beyond the limitations of contemporary language. It is here at the edge of language that poetry is retrieved from the void. A.I. will likely absorb more and more of what the culture has digitized of its literature, and yet, A.I.’s ability to project outward and beyond these models into startling new works of vital originality seems, at best, uncertain. Poet Ezra Pound said that “artists are the antennae of the race,” and by race we can infer the term culture. What Pound implies by this statement is that there is a human responsibility on the part of the artist to see beyond all limitations. Despite the promise of A.I. technology as a creative tool, this seeing beyond requires the elemental poetic vision of subconscious dream-song, sound-sense, and purest imagination.


December 16, 2022 – January 18, 2022


[1] L’imagination est la seule arme dans la guerre contre la réalité.” [Imagination is the only weapon in the war against reality.] Jules de Gaultier (1858–1942) was a French philosopher and essayist.

[2] []

[4] Ibid.

[5] Not to be outdone, in October of 2021, Aum Golly: Poems on Humanity by an Artificial Intelligence was published. Assisted by typist Jukka Aalho, “Aum Golly is a [72-page] book of [50] poems written in 24 hours. It was made possible by GPT-3 – an advanced autoregressive language model published in 2020 by OpenAI.”

[7] Midjourney is an independent research lab that produces a proprietary artificial intelligence program that creates images from textual descriptions, similar to OpenAI's DALL-E and the open-source Stable Diffusion. The tool is currently in open beta, which it entered on July 12, 2022. Source: Wikipedia.

[10] From the back cover of the Granary Books 2001 edition of I Remember. See also,

[13] rob mclennan, from “Selecting Phil Hall,” Jacket2, January 1, 2015:

[14] For a deeper investigation into the potentialities of contemporary poetry see my own work in Part Thief, Part Carpenter, (Beautiful Outlaw Press, 2021).









Mark Goldstein is the author of five books including the award-winning Her Process (Beautiful Outlaw, 2021), and Form of Forms (BookThug, 2012). Goldstein is a writer, translator, and editor whose poetry and criticism has appeared in periodicals and journals such as The Capilano Review, Periodicities, and Jacket2. He is also a graphic artist, musician, and the founder of Beautiful Outlaw Press. He lives in Toronto.

Thanks to Randall Baker, Phil Hall, Karen Hill and Jaclyn Piudik for their feedback on this essay.


most popular posts